Andrew Hoopes

Results 35 comments of Andrew Hoopes

Right. Each line in the 'pairs' text file should define a (moving, fixed) image pair to align, for example: ``` OASIS_OAS1_0001_MR1 OASIS_OAS1_0002_MR1 OASIS_OAS1_0003_MR1 OASIS_OAS1_0004_MR1 ... ``` Then you could, for...

@WwwwJX which paper are you referring to? While you might have only one fixed image for atlas-based (subject-to-atlas) registration models, you can use any arbitrary image pair at test-time for...

We will be providing new modes for the redesign branch. The current models are probably not loadable since the default net architecture has changed a bit

Where did you get `model.h5` from? Is it a `VxmDense` model? In the refactored master branch, the default model configurations have changed a bit, and probably won't be able to...

Just an update on this - voxelmorph 0.2 has now been released on pypi. This version includes the unet_half_res argument from the development branch (among other changes 😃)

yes, new pytorch model weights will be added soon

Hi @haru121498. To debug this, we will need some info about the exact command you ran as well as where you got the code from (e.g. github dev/master or pip...

The command line options provided to train.py would be most useful here

You'll need to provide the appropriate command line arguments described in the *training* section of the vxm README: https://github.com/voxelmorph/voxelmorph/blob/dev/README.md

Correct, setting `--int-steps 0` will disable diffeomorphisms